Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Bias in random forest variable importance measures: illustrations, sources and a solution.

Carolin Strobl1, Anne-Laure Boulesteix, Achim Zeileis

  • 1Institut für Statistik, Ludwig-Maximilians-Universität München, Ludwigstr, 33, 80539 München, Germany. carolin.strobl@stat.uni-muenchen.de

BMC Bioinformatics
|January 27, 2007
PubMed
Summary

Random forest variable importance measures can be unreliable for selecting important variables when data types vary. An improved random forest method offers unbiased variable selection for genomics and bioinformatics research.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Transformation Discriminant Analysis for Constructing Optimal Biomarker Combinations.

Statistics in medicine·2026
Same author

When to Adjust for Multiple Testing: A Unifying Guiding Principle.

Biometrical journal. Biometrische Zeitschrift·2026
Same author

Likelihood-based modeling of covariate-specific time-dependent receiver operating characteristic curves.

Statistical methods in medical research·2026
Same author

Rejoinder to the discussion on ''Nonparanormal Adjusted Marginal Inference''.

Biometrics·2026
Same author

Nonparanormal adjusted marginal inference.

Biometrics·2026
Same author

Smooth transformation models for survival analysis: A tutorial using R.

Statistical methods in medical research·2026

Area of Science:

  • Bioinformatics
  • Computational Biology
  • Genomics

Background:

  • Random forests are widely used for variable selection in bioinformatics.
  • Variable importance measures are crucial for identifying relevant genetic markers for disease prediction.
  • Current methods are unreliable when predictor variables have varying scales or categories.

Purpose of the Study:

  • To evaluate the reliability of random forest variable importance measures.
  • To address limitations in variable selection for diverse data types in bioinformatics.
  • To propose an improved random forest implementation for unbiased variable selection.

Main Methods:

  • Simulation studies were conducted to assess variable importance measures.
  • An alternative random forest implementation was developed.

Related Experiment Videos

  • Subsampling without replacement was employed in the proposed method.

    • Main Results:

    • Standard random forest variable importance measures yield misleading results with mixed data types.
    • Biased selection in individual trees and bootstrap sampling contribute to inaccuracies.
    • The proposed method demonstrates unbiased variable selection across different data scales and categories.

    Conclusions:

    • The novel random forest implementation ensures reliable variable selection for heterogeneous data.
    • This approach is suitable for genomics and computational biology applications.
    • The method is readily applicable in R for bioinformatics research, as demonstrated with RNA editing data.